International Refereed Journal of Engineering and Science (IRJES)

ISSN (Online) 2319-183X, (Print) 2319-1821

Volume 4, Issue 3 (March 2015), PP.01-07

PSoC BASED SPEECH RECOGNITION SYSTEM

1

Shubhasini Sugumaran, 2Mr.V.R.Prakash

Department of Electronics and Communication Hindustan University Chennai, India

Department of Electronics and Communication Hindustan University Chennai,India

Abstract: Speech Recognition Systems(SRS) have been implemented by various processors including the
digital signal processors(DSPs) and field programmable gate arrays(FPGAs) and their performance has been
reported in literature. The fundamental purpose of speech is communication, i.e., the transmission of
messages.In the case of speech, the fundamental analog form of the message is an acoustic waveform, which we
call the speech signal. Speech signals can be converted to an electrical waveform by a microphone, further
manipulated by both analog and digital signal processing, and then converted back to acoustic form by a
loudspeaker, a telephone handset or headphone, as desired.The recognition of speech requires feature extraction
and classification. The systems that use speech as input require a microcontroller to carry out the desired
actions. In this paper, Cypress Programmable System on Chip (PSoC) has been studied and used for
implementation of SRS. From all the available PSoCs, PSoC5 containing ARM Cortex-M3 as its CPU is used.
The noise signals are firstly nullified from the speech signals using LogMMSE filtering. These signals are then
sent to the PSoC5 wherein the speech is recognized and desired actions are performed.
Keywords: PSoC, LogMMSE, Speech Recognition

I.

INTRODUCTION

The basic idea of speech is the transmission of messages. A message is represented as a sequence of
discrete symbols that quantifies its information in bits and the rate at which information is transmitted as bits per
second (bps). Speech recognition techniques have seemed to be more efficient and convenient for humanmachine interaction. The speech recognition systems with fixed vocabulary were deployed in many
applications[8] and [9]. Speech recognition systems for voice operated application have been implemented using
various hardware platforms such as the DSPs [3], FPGAs [5] and microprocessors [10]. In speech production,
the information to be transmitted is encoded in the form of a continuously varying analog waveform that can be
transmitted, recorded, manipulated, and ultimately decoded by a human listener. This analog signal is the speech
signal. These signals tend to be corrupted by noise in the real world. If the noise can be estimated from the noise
source, this estimated noise can then be subtracted from the primary channel resulting in the desired signal. This
task is usually done by linear filtering. In real time situations, the corrupting noise is a nonlinear distortion
version of the source noise, so a nonlinear filter should be a better choice. To reduce the influence of noise in the
speech, speech enhancement is done. The recognition algorithm without the use of enhancement algorithms
proved to be less efficient.
Programmable System on Chip (PSoC) have and are being employed in a number of applications. They
are cost effective due to which they have limited storage and computational power. In context to this, the
recognition accuracy becomes important for PSoC and is addressed in the paper.

II.

PROGRAMMABLE SYSTEM ON CHIP (PSOC)

Programmable System on Chip(PSoC) has been designed and implemented by Cypress semiconductors
[2] and [4]. Every PSoC contains a microcontroller, programmable analog blocks such as ADC, DAC, I/O
drivers and digital blocks such as Universal Digital Blocks (UDBs), CAN, I2C, PWM in a single chip.
Embedded Development kits from Cypress contain one of the three PSoCs PsoC1, PSoC3 and PSoC5. The
processing performance, functionality, internal memories and configurability of the PSoC increases from
PSoC1through PSoc5.

www.irjes.com

1 | Page

PSoC BASED SPEECH RECOGNITION SYSTEM

FEATURES
CPU
Flash
Interface

ADCs
DACs
I/Os

PSoC1
8-bit M8C
core
4 to 32kB
IC,
SPI,
UART,
FS USB 2.0

1
deltasigma
2(6 bit)
64

PSoC3
8-bit 8051
8 to 64kB
IC,
SPI,
UART,
LIN,
FS
USB
2.0,
IS,
CAN
1
DeltaSigma
4(8 bit)
72

PSoC5
32-bit ARM
Cortex
32 to 256kB
IC,
SPI,
UART, LIN,
FS USB 2.0,
IS,CAN

1 Delta-Sigma,
2SARs
4(8 bit)
72

Table 1: Comparison of PSoCs

PSoC5 uses PSoC Creator for its development and implementation. PSoC Creator is a visual
development tool and Integrated Development Environment for PSoC. It has a rich library of prebuilt
components and a schematic design entry tool. It combines C based development flow with an automatically
generated Application programmable interface (API). API rduces the errors in code and ensures proper
interfacing with the peripheral devices which enables the software development to be faster, easier and less
prone to errors. The PSoC Creator also has powerful, modern debugger, which is built in the IDE. It is used to
display the values after execution at each point.

III.

SPEECH RECOGNITION SYSTEM

The block diagram is as shown in Fig.1 below.

Speech Recognition System

The speech is given as input through a microphone. The given speech might contain external
disturbances called noise. This noise is cancelled using LogMMSE algorithm. The speech signal after noise
cancellation is amplified and filtered to remove further disturbances in the signal. The speech signal in analog
form is then converted into digital signal using inbuilt analog to digital converter of PSoC. The feature
extraction and pattern recognition is done using MFCC algorithm. The analog to digital conversion is done
using delta sigma ADC present in PSoC5.
The sampling rate of ADC can be adjusted depending on the speech signal as required. The delta sigma
ADC contains three blocks an amplifier, a modulator and a decimator. The decimator is a four stage CIC
decimation filter. It also contains a post processing unit. Continuous mode of the ADC is being used for
conversion. The output of the speech recognition system is shown through the movement of a robot in different
directions.

www.irjes.com

2 | Page

PSoC BASED SPEECH RECOGNITION SYSTEM

IV.

TECHNIQUES USED

The performance of the SRS degrades when implemented in real world environment. This degradation
is due to acoustic model mismatch. The acoustic model mismatch describes the difference between the
environment in which the SRS is tested and the actual environment in which it is deployed. This can include
echoes, background noise, speaker variability and transmission effects.
1.

Log MMSE Filtering

Fig.2 Noise reduction using Log MMSE

1.1 Preprocessing
In preprocessing of audio signals start with pre-emphasis refers to a system process designed to
increase the magnitude of some frequencies with respect to the magnitude of other frequencies in order to
improve the overall signal-to-noise ratio by minimizing the adverse effects of such phenomena as attenuation
distortion or saturation of recording media in subsequent parts of the system. The mirror operation is called deemphasis, and the system as a whole is called emphasis.
Pre-emphasis is achieved with a pre-emphasis network which is essentially a calibrated filter. This
network composed of two resistors and one capacitor. The frequency response is decided by special time
constants. The cutoff frequency can be calculated from that value. Pre-emphasis is commonly used
in telecommunications, digital audio recording, record cutting, in FM broadcasting transmissions, and in
displaying the spectrograms of speech signals.
De-emphasis is the complement of pre-emphasis, in the anti noise system called emphasis. Emphasis is
a system process designed to decrease, (within a band of frequencies), the magnitude of some (usually higher)
frequencies with respect to the magnitude of other (usually lower) frequencies in order to improve the overall
signal-to-noise ratio by minimizing the adverse effects of such phenomena as attenuation differences or
saturation of recording media in subsequent parts of the system.
1.2 Segment
The signal samples are segmented into fixed number of frames and each frame samples are evaluated with
hamming window coefficients.
The total frames are calculated by,
Fn = (Ls Ns)/ (Ns*Sp) +1
Where,
Ls = length of signal
Ns = Length of each frame
Sp = Shift Percentage
Finally the samples of each frames are separated from input signal using Fn and Sp and its scaled by the
hamming window coefficients.

www.irjes.com

3 | Page

PSoC BASED SPEECH RECOGNITION SYSTEM

0.06
0.7

0.04

0.6

0.5

0.02

0.4

0
0.3

-0.02
0.2

-0.04
0.1

-0.06

20

40

60

80

100

120

140

160

180

200

20

40

60

80

100

120

Fig.3 Segmented signal

1.3 Filtering
After the signal segmentation, the magnitude and phase spectrum from noisy signal are computed by
applying fast fourier transform.The magnitude of noisy signal spectrum is further utilized for filtering process
and signal phase kept same.
The restored signal magnitude spectra is obtained by,
Rs = G .* Y
Where, G Log spectral amplitude Gain function
Y Magnitude response of noisy signal
The log spectral gain function is defined by,
G = x./(1+x) exp(eint (v))
Where, v = x./(1+x) * r
x and r Priori and posteriori signal to noise ratio
eint Exponential integral
The posteriori snr is defined by,
r = (Y.^2)/ lamda
lamda = E[(Y).^2]
Where, lamda - Noise power spectrum variance
E Mean value
Complex spectrogram obtained by Filtered magnitude spectrum is combined with noisy signal phase spectrum.
The restored signal is reconstructed by applying inverse fast fourier transform to this complex spectrogram.The
performance of filtering is measured with SNR evaluation and it is defined by,
SNR= 10log10 (Msig^2/(sum(in-out)^2/Ls))
Where, Msig = Maximum amplitude of signal
in, out= Noisy input signal and restored output.
1.4 Concatenate
The signals that were obtained on segmentation are now filtered and the noise content is reduced. These signals
are then concatenated to reform the original signal.
1.5 Restore
The restored noiseless signal is restored to its original form.
2. MEL FREQUENCY CEPSTRUM COEFFICIENT(MFCC)
To increase the robustness in the frame selection process, a robust feature extraction with short time
Fourier transform (STFT) domain uncertainty propagation (STFT-UP) was used. Our implementation followed
using STFT-UP to compute a minimum mean square error (MMSE) estimate directly in the mel-frequency
cepstral coefficient (MFCC) domain. For this particular implementation, amplitude based MFCCs with cepstral
mean subtraction were used to attain improved performance.

www.irjes.com

4 | Page

PSoC BASED SPEECH RECOGNITION SYSTEM

Fig. 4 MFCC
In speech recognition, the Mel-frequency cepstrum (MFC) is a representation of the shortterm power
spectrum of a sound, based on a linear cosine transform of a log power spectrum on a nonlinear Mel scale of
frequency.
MFCC = DCT [ LOG [ ABS [ FFT (SPEECH) ]]]
Mel-frequency cepstral coefficients (MFCCs) are coefficients that collectively make up an MFC. The difference
between the cepstrum and the mel-frequency cepstrum is that in the MFC, the frequency bands are equally
spaced on the mel scale, which approximates the human auditory system's response more closely than the
linearly spaced frequency bands used in the normal cepstrum. This frequency warping can allow for better
representation of sound.

V.

TRAINING AND TESTING

The goal of the system training and inventory design stage is twofold: we need to divide the inventory
into collections of phonetically similar segments with varying lengths and we need to arrive at a statistical
description that tells us which set of collections is most likely to contain the inventory subsection that best
matches the underlying clean frame of an incoming noisy frame . The division of the inventory into the
collections is performed in a step-by-step fashion. First, is segmented and all silent segments are removed. The
non-silent part of the inventory is then divided into sections that each belong to one of 40 phonetic classes.
We are applying the feature extraction to the entire segment stream of the inventory. Because the inventory
signal is assumed to be virtually undistorted, it is sufficient to only retain the resulting short-time MFCC feature
means and to discard the associated variance estimates. The feature means become, thereby, essentially feature
vectors in their own right and we can develop a cluster model for them. We have decided to use the means and
not the actual cepstral vectors at this stage to ensure that the impact of the mean-extraction-processing is
captured in our feature representation.

VI.

RESULTS

Fig. 5 Noisy Signal

www.irjes.com

5 | Page

PSoC BASED SPEECH RECOGNITION SYSTEM

Fig. 6 Log MMSE Filtering

Fig. 7 MFCC Feature Extraction

For experimental basis, speech signals from noisy environments were taken. The result shown is fr a
signal in the airport. The noise is cancelled using LogMMSE Filtering and feature extraction is done by MFCC.

VI.

CONCLUSION

In this paper speech recognition is done and appliances are controlled using speech commands. The
commands are given by the user in the form of speech. These signals are filtered using LogMMSE filtering due
to which environmental noise is nullified. Further as a part of feature extraction, MFCC is used. A database is
created where all the commands are saved. These commands are then given as input to the PSoC5 kit where the
PSoC is programmed to give the desired result. According to the PSoC, the robot connected to the PSoC moves
in different directions as specified.

REFERENCES
[1].

[2].

[3].

[4].

V. Naresh, B. Venkataramani, Abhishek Karan and J. Manikandan, " PSoCbased isolated speech
recognition system, " International conference on Communication and Signal Processing, April 3-5,
2013.
R Namba, K Kobayashi, T Ohkubo and Y.Kurihara, "Development of PSoC microcontroller based
solar energy storage system," Proceedings of SICE Annual Conference (SICE), 2011, pp.718-721,
2011.
J. Manikandan, B. Venkataramani, K. Girish, H. Karthic, V. Siddharth, "Hardware Implementation of
Real-Time Speech Recognition System Using TMS320C6713 DSP",24th International Conference
onVLSI Design (VLSI Design),pp.250-255, 2011
Jingchuan Wang and WeidongChen , "Integration of PSoC technology with educational robotics",
International Conference on Field-Programmable Technology (FPT), 2010, pp.332-336, 2010.

www.irjes.com

6 | Page

PSoC BASED SPEECH RECOGNITION SYSTEM

[5].

Cheng-Yuan Chang , Ching-Fa Chen , Shing-Tai Pan , Xu-Yu Li, " The speech recognition chip
implementation on FPGA , Mechanical and Electronics Engineering (ICMEE), 2nd International
Conference,2010.
[6]. V. Amudha, B. Venkataramani, J. Manikandan, "FPGA implementation of isolated digit recognition
system using modified back propagation algorithm,"International Conference on Electronic Design
ICED 2008, pp.1-6.
[7]. V.Amudha, B.Venkataramani, R.Vinoth Kumar and S. Ravishankar,SOC Implementation of HMM
Based Speaker Independent Isolated Digit Recognition System, in Proc. of IEEE Int. Conf. on
VLSIDesign VLSI07, 2007, pp.848-853.
[8]. Trihandyo, A. Belloum, A. Kun-Mean Hou, A real-time speech recognition architecture for a multichannel interactive voice response system, Iternational Conference on Acoustics, Speech, and Signal
ProcessingICASSP-97, vol.2, pp.1527-1530,1997
[9]. Mike Wald, Using Automatic Speech Recognition to Enhance Education for All Students: Turning a
Vision into Reality [A].In Proceedings of 34th ASEE/IEEE Frontiers in Education Conference S3G,
Indianapolis, Indiana, 2005, pp 22-25.
[10]. N. Hataoka, H. Kokubo, Y. Obuchi, and A. Amano, "Compact and robust speech recognition for
embedded use on microprocessors," IEEE Workshop on Multimedia Signal Processing, pp. 288-291, 911 Dec. 2002.

www.irjes.com

7 | Page